Commonly Interesting Images

Abstract

Images tell stories, trigger emotions, and let us recall memories -- they make us think. Thus, they have the ability to attract and hold one's attention, which is the definition of being "interesting". Yet, the appeal of an image is highly subjective. Looking at the image of my son taking his first steps will always bring me back to this emotional moment, while it is just a blurry, quickly taken snapshot to most others. Preferences vary widely: some adore cats, others are dog enthusiasts, and a third group may not be fond of either. We argue that every image can be interesting to a particular observer under certain circumstances. This work particularly emphasizes subjective preferences. However, our analysis of 2.5k image collections from diverse users of the photo-sharing platform Flickr reveals that specific image characteristics make them commonly more interesting. For instance, images, including professionally taken landscapes, appeal broadly due to their aesthetic qualities. In contrast, subjectively interesting images, such as those depicting personal or niche community events, resonate on a more individual level, often evoking personal memories and emotions.

Figures (8)

• Figure 1: DALL·E 3: "Give me an image which shows the contrast between interesting and uninteresting [...]" dalle32023interesting. In the generated image, many colors are used for the interesting part, which is usually overfilled with objects and is generally quite complex. However, the uninteresting part is usually depicted with few objects, monotonous and colorless, which gives a tendency towards simplicity.
• Figure 2: The image space is partitioned into semantic similar partitions using k-means and hierarchical clustering (a). Images from users are assigned to their respective partitions (b). The more users have liked an image in a particular partition, the more commonly interesting it is considered (c). All partitions, $CI$ scores, and typical images are included in the supplementary material for a closer investigation.
• Figure 3: Example of a merged cluster: Around 60 percent of users who prefer cat images in partition $p_{37}$ also prefer dog images in partition $p_{5}$. As the semantic similarity $d(p_{37},p_{5}) = 1.7$ is close enough these two partitions get merged into $p_{37.5}$.
• Figure 4: $CI$ scores overlaid on the image partitions space (a) and correlation with the number of images in each partition (b).
• Figure 5: Meta-data such as the absolute number of views and likes as metrics are unreliable due to recommendation system influences. An image with a few likes or views can also be interesting (a), while viral images might be irrelevant to many users (b).